Professional Documents
Culture Documents
+ =
=
[
=
=
1
2
1
1
2
1
2
1
e
i j
j
j i
j
e
i
N
N
( 2 )
By adding a third (center) node, the same technique can be used to obtain a quadratic
interpolation (Equation 3).
| | ( )( ) ( )( ) | |
2
1 1
2
1
1
2
1
+ =
e
N ( 3 )
Up to this point, with the Lagrange approach, if we desired to increase the interpolation order, we
need to define totally new shape functions. With the hierarchic approach, to get new functions, we
simply add some terms to the old functions. Thus, we add some internal interpolation functions,
i.e. edge functions
e
N
A
, to nodal shape functions
e
N
N
(Equation 4).
| | | |
e
A
e
N
e
N N N = ( 4 )
As the element square matrix will always involve an integral of the product of the derivatives of
the interpolation functions, hierarchical functions are based on orthogonal polynomials. Thats
why p-FEM is generally implemented using Legendre polynomials defined by Equation 5 called
Rodrigues formula.
| |
( ) ( )
(
=
> e e
n
n
n
n
n x
dx
d
n
x L
n / n x
1
! 2
1
, 2 N , 1 ; 1
2 ( 5 )
In the one-dimensional case, to obtain a 2 > p order of interpolation using the hierarchical
approach, we can use Equation 6 and Equation 7.
2009 SIMULIA Customer Conference 5
| | ( ) ( )
{ }
( ) | |
e
p
e
N N
A1
1
2
1
1
2
1
+ = ( 6 )
{ }
( ) ( ) ( ) | | 2
A1
2 4
1
, 2 / N
=
> e
p p
e
p
L L
p
N
p p
( 7 )
In the same way, it is easy to construct hierarchical interpolation functions for 2D and 3D
elements. As we said earlier, for a one-dimensional element, we have nodal and edge interpolation
functions: nodal functions are defined by Lagrange polynomials and edge functions by Legendre
polynomials. For a two-dimensional element, we must add face functions
e
N
F
(Equation 8) and
for a three-dimensional element internal functions
e
N
I
(Equation 9). Face functions and internal
functions are respectively defined by a combination of ( ) q , interpolations and ( ) , q , ,
interpolations.
| | | |
e e e e
N N N N
F A N
= ( 8 )
| | | |
e e e e e
N N N N N
I F A N
= ( 9 )
The Figure 2 gives a graphical representation of a bilinear nodal interpolation associated with a
2nd order hierarchical interpolation for a four nodes 2D element. We can see the representation of
four nodal functions, four edge functions and one face function. For the same element, the Figure
3 gives a graphical representation in the case of a 3rd order of interpolation.
6 2009 SIMULIA Customer Conference
Figure 2. 2D, four nodes element, 2nd order interpolation: graphical representation
of four nodal functions (bilinear), four edge functions and one face function.
Figure 3. 2D, four nodes element, 3rd order interpolation: graphical representation
of a four nodal functions (bilinear), four edge functions of degree two, four edge
functions of degree 3 and four face functions (degree 2x2, 3x2, 2x3, 3x3).
2009 SIMULIA Customer Conference 7
Up to now, we have shown isotropic increase of the element order. With the hierarchical
approach presented above, it is possible to adopt an anisotropic p-refinement of the element
(Dster, 2001). Indeed, we can define independent degrees of interpolation for each space
directions. An example can be seen on Figure 4.
Figure 4. 2D, four nodes element, 5th order interpolation for one edge of the
element.
2.2 A specific formulation for capturing edge effects
Capture the strong variations of the solution in the vicinity of the ends of the joint For this goal
and to introduce more easily the formulation suggested, let us consider a one-dimensional
academic problem: an ideal straight bar composed of a linearly elastic material and subjected
to an axial tension load as depicted in Figure 5. The solution ( ) x u x describes the
displacement of the bar in x direction. The bar is loaded by a force per unit length ( ) x f x in
the bar cross-sectional direction and a punctual load F for = x . The Young modulus is
denoted E , the cross-sectional area A, the density and the bar length . The problem is
modeled by Equation 10 with a clamped end for 0 = x .
8 2009 SIMULIA Customer Conference
Figure 5. The academic bar problem: a simple one dimensional case to introduce
the formulation of the user element family.
( ) ( )
( )
= = =
(
= = =
O e = +
x
A E
F
dx
du
x u
f A
dx
u d
A E
L
, L M for
0 , O M for 0 O
M 0
2
2
( 10 )
Imposing a solution with a very strong variation, we study the efficiency of the hierarchical
formulation according to the selected degree. In Figure 6 the approximation for a degree 8
hierarchical interpolation is plotted together with the exact solution of the problem (Equation 11).
( )
50
x x uex = ( 11 )
At first sight, with a single element of degree 8, we manage with difficulty to approach the exact
solution of degree 50. We can of course continue to increase the order of interpolation or refine
the mesh but that would be to the detriment of the computing time and the aims of this work. It
also should be noted that the oscillations observed do not facilitate the research of the result. If the
integration of the stiffness matrix and the elementary vectors is accessible here analytically, that is
seldom the case for most complex problems. When the element is deformed, the geometrical
transformation is not linear anymore and the J acobian matrix is a sophisticated polynomial
function. Explicit integration becomes impossible and it is then necessary to use numerical
integration techniques.
In this paper, we suggested a way to improve the approximation and capture edge effects without
increase the order of interpolation or refine the mesh (Joanns, 2007).
2009 SIMULIA Customer Conference 9
Figure 6. Approximation of the exact solution by a degree 8 hierarchical
interpolation. Integration points are represented by square symbols.
In order to attenuate oscillations phenomena and better capturing edge effects, we should optimize
the element previously described and distend the solution near the element boundaries. We
choose to supplement the linear geometrical interpolation (Equation 12) by a suitable cubic
interpolation. It is essential that this new function dont modify the values on, but near the
boundaries.
| | ( ) ( )
(
+ = 1
2
1
1
2
1
e
P ( 12 )
The distending parameter is denoted o and we define a new function ( ) o o , , A which
is the cubic additional polynomial (Equation 13).
| | ( ) ( ) ( ) ( )
(
+ = o o , 1
2
1
, 1
2
1
A A P
e
( 13 )
10 2009 SIMULIA Customer Conference
If 1 = o there is no additional transformation but when 0 o the distending becomes
maximal. The new geometrical interpolation functions are no longer identical to the nodal
interpolation of the solution, except if 1 = o . We choose to preserve a pseudo-parametric or sub-
parametric element and the additional transformation intervenes only when 3 > p .
While preserving the degree 8, this additional geometrical transformation enables us to find a
value of o for which the error of approximation becomes minimal (Figure 7).
Figure 7. Approximation of the exact solution by a degree 8 hierarchical
interpolation and applying a geometrical distending of the solution near the
boundaries of the element.
3. A suitable CAE methodology
The family of elements discuss previously is implemented as an Abaqus user element and is
associated with a dedicated pre- and post-processing (Abaqus Python scripts). With this
methodology, it is possible to preserve industrial meshes with a reasonable description of the
adhesive joint behavior while ensuring a total compatibility with Abaqus code. This tool takes the
form of a user subroutine which allows enriching a portion of the mesh, for example the
adhesive layer, with a hierarchical formulation. The subroutine is organized in a modular way and
allows managing the elementary matrix and vectors construction. Since we act only on specific
zones, the global assembly as well as the resolution of the problem is done by Abaqus. In
complement to the UEL subroutine, we have developed pre- and post-processing capabilities. The
purpose of the pre-processing is to transform the input file describing the problem (mesh,
boundary conditions and loads) in order to substitute the standard elements of the adhesive zone
by elements with hierarchical formulation. The program integrates a strategy of classification of
the additional degrees of freedom particularly efficient: the enrichment of hierarchical elements is
entirely controlled by the edges of the elements. Coupled with an error indicator, this choice
makes it possible to consider automatic calculations.
2009 SIMULIA Customer Conference 11
Post-processing provides graphical information for Abaqus visualization module by representing
the results obtained with hierarchical elements. Taken as a whole, the tool implemented represents
about ten thousand lines of Fortran and Python codes with nearly two thirds devoted to the central
subroutine (UEL). The implementation presented above allows considering three dimensional
structural problems. It is based on a hexahedral element formulation, being able to vary the
polynomial degree for the three local directions. The use of anisotropic interpolation leads to very
efficient approximations.
3.1 Patch tests
The successful application of finite element analysis should always include a validation of the
elements to be used and a validation of their implementation in a specific computer program. A
single element test can show the effects of the element geometrical parameters such as convexity,
aspect ratio or skewness. The patch test has been introduced as a complementary check for
continuity intra- and inter-elements. It was developed from physical intuition and later written in
mathematical forms. The concept is fairly simple: to be assured of convergence one must be able
to exactly satisfy the state where the derivatives, in the governing integral statement, take on
constant or zero values. Thus, the patch test provides a simple numerical way to verify that correct
programming was achieved and that the element behaves as it should. We define a patch of
elements to be a mesh where at least one node is completely surrounded by elements. Any node of
this type is referred to as an interior node. The outer nodes are referred to as exterior nodes. The
exterior nodes are utilized to introduce the essential boundary conditions and loads required by the
test. The Figure 8 presents the patch meshes used for 2D and 3D user elements. In order to verify
the transition between standard and user elements, exterior elements are CPE4 or C3D8 elements
and share at least one node with user inner elements. By moving the central node, we have
distorted user elements to see the effect that has on the numerical accuracy of the patch. With
various interpolation configurations, we have applied rigid body modes, constant strain and
constant stress state and controlled the accuracy of the patch response. For example, a rigid body
displacement of the patch should not introduce strain nor stress change.
Figure 8. Patch meshes for 2D and 3D user elements tests.
12 2009 SIMULIA Customer Conference
3.2 2D validation example
Previous tests have validated the behavior of a family of user elements (2D and 3D) based on a
hierarchical interpolation and implemented in Abaqus (Abaqus, 2006). The following paragraph
highlights the relevance of the methodology for high stress gradients. The problem presented here
is directly derived from the 2D patch-test: by the removal of four elements and the readjustment of
the central node, we obtain a L shape. This configuration is particularly interesting from a
numerical point of view since it induces a very strong gradient of stress. The model comprises
twelve 2D elements with four geometrical nodes. The three central elements have elastic
properties ten times higher than those of the peripheral elements. The test consists in imposing a
uniform displacement on the higher border of the L, while immobilizing in a suitable way the base
and the side (Figure 9). On the peripheral edges, the degree of interpolation is fixed (linear
interpolation). For interior edges, the degree is free to evolve anisotropically. A reference mesh,
comprising 2700 elements with linear interpolation (5642 degrees of freedom), makes it possible
to assess the relevance of the hierarchical model which comprises only 378 degrees of freedom on
the Figure 10 (with a maximum degree of four in that case). The Figure 10 confronts the results
obtained for the shear stress field. We can note a very good correlation of the results, even in re-
entrant corner zone while having here very little elements and a relatively weak interpolation
degree (only four).
Figure 9. The L shape helps the numerical validation of the hierarchical
interpolation with very strong gradients of stress. Representation of the boundary
conditions and visualization of the displacement field (magnitude) induced.
2009 SIMULIA Customer Conference 13
Figure 10. Confrontation of the results obtained on a benchmark model (left: 2700
bi-linear elements) and by a hierarchical approach (right: twelve user elements).
The good correlation of these results shows the relevance of the hierarchical
approach for capturing strong variations of the solution.
3.3 3D validation example
In addition to the 2D validation problem, we wished to work on three-dimensional subsystems
representative of adhesive bonded joints. Three configurations were defined: single lap shear
test (Figure 11), peeling test (Figure 12) and compression on tubes. For each one of these tests, we
used the hierarchical approach combined with a non-linear anisotropic behavior dedicated to
adhesive exhibiting high hydrostatic sensitivity (J oanns, 2007).
Starting from a conventional input file, the pre-processing replaces the standard elements of the
targeted zones by user elements with hierarchical formulation. The interpolation compatibility on
transition zones is assured. The initial degree is fixed beforehand but can evolve locally according
to the need. This local adaptation could be obtained automatically by using an error indicator
related to each edge. During this adaptation phase, it is important to use the simplest behavior
model for the adhesive (linear elasticity). Once the desired precision reached, calculation can be
launched with a non-linear and anisotropic behavior. The principal subroutine generates results
files for nodal variables and integration points evaluation. Post-processing performs data
conversions for the visualization module. It is then possible to display results within the adhesive
bonded joint (Figure 12).
14 2009 SIMULIA Customer Conference
Figure 11. Single lap joint. Elements with standard formulation are replaced by
elements with hierarchical formulation around the adhesive joint; the interpolation
compatibility on the transition zone is assured.
Figure 12. Peeling test. Post-processing performs data conversions for the
visualization module. It is then possible to display results obtained by the
hierarchical approach within the adhesive bonded joint.
4. Concluding remarks
The use of finite element methods allow the study of complex configurations. Nevertheless, the
degree of accuracy is in direct correspondence with the number and the type of elements used.
Very often, localized stresses in multi-materials structures impose the use of a great number of
2009 SIMULIA Customer Conference 15
elements. Developments of the numerical analysis led to the emergence of specific tools which
remain more or less applicable to the targeted industrial cases. A new approach, based on a
hierarchical formulation, has been presented in this paper. This tool takes into account the
localized phenomena governing the behavior of adhesive joints while guaranteeing compatibility
with a coarser industrial modeling. This work brings an appropriate response to industrial
constraints. Confrontations between experimental tests and numerical calculations are particularly
encouraging. Finally, the modularity of the implementation allows its extension to other fields of
applications such as detecting delamination on composites materials.
5. References
1. Abaqus/Standard, User's Manual, version 6.6, Hibbitt, Karlsson and Sorensen Inc, 2006.
2. Adams, R. D. and Mallick, V., A method for the stress analysis of lap joints, The J ournal of
Adhesion, vol. 38, no.3-4, pp. 199-217, 1992.
3. Babuka, I. and Szab, B. A., Finite element analysis, J ohn Wiley and Sons, 1991.
4. Besson, J., Cailletaud, G., Chaboche, J . L. and Forest, S., Mcanique non linaire des
matriaux, Herms, 2001.
5. Bigwood, D. A. and Crocombe, A. D., Non-linear adhesive bonded joint design analysis,
International Journal of Adhesion and Adhesives, vol. 10, no. 1, pp. 31-41, 1990.
6. Chaboche, J. L., Feyel, F. and Monerie, Y., Interface Debonding Models: A viscous
regularization with a limited rate dependency, International J ournal of Solids and Structures,
vol. 38, no. 18, pp. 3127-3160, 2001.
7. Cugnon, F. Automatisation des calculs lments finis dans le cadre de la mthode-p, Ph. D.
Thesis, Universit de Lige, Belgium, 2000.
8. Dean, G. D. and Crocker, L., The Use of Finite Element Methods for Design with
Adhesives, Measurement Good Practice Guide, no. 48, National Physical Laboratory, UK,
2001.
9. Dster, A., High order finite elements for three dimensional, thinwalled nonlinear continua,
Ph. D. Thesis, Technische Universitt Mnchen, Germany, 2001.
10. Fays, S., Adhesive Bonding Technology in the Automotive Industry, Adhesion and
Interface, vol. 4, no. 2, pp. 37-48, 2003.
11. J oanns, S., Caractrisation mcanique et outil daide au dimensionnement des collages
structuraux, Ph. D. Thesis, Ecole Nationale Suprieure des Mines de Paris, ParisTech,
France, 2007.
12. Peano, A. G., Hierarchies of conforming finite elements, Ph. D. Thesis, Washington
University, St. Louis, USA, 1975.
13. Zienkiewicz, O. C., Taylor, R. L. and Zhu, J . Z., The Finite Element Method, its basis &
fundamentals, Elsevier Butterworth-Heinemann, Sixth Edition, 2005.